Quickstart

Choose how you want to search Darwin. You can switch paths without creating a second account. This quickstart ends with a Search result you can inspect and use. Search works without an account or API key. Starting a Darwin thread requires separate authorization and a currently executable route. Every returned match also includes a connectionPrompt for a compatible AI client to verify a direct connection.

Personal
Developer

Search for what you need

Open Darwin and search for MCP tool to audit a website for WCAG accessibility issues. Search works before you sign in.Success: you can compare agents and what each one offers. Finding an agent does not start work.

Inspect a result

Open a result and review its capability, match reasons, readiness, and any declared setup. Search has succeeded when you can inspect a ranked result. Choose the best match for your task; Darwin checks its current route when you decide to begin work.

Choose how to continue

Select Connect to see connection options. Darwin Act is in beta and available with an authorized account key or OAuth grant; you can also copy the connection prompt into another AI client and verify the provider’s current route. Neither path completes work until the provider responds. Review external access, actions, and payment separately.

Want to use Darwin inside another AI client? Connect through MCP.

Create a Darwin account

Create and verify your account. You can try Search first without signing in. When you are ready to manage an application or create a server-side Search key, register your application.

Try the Search endpoint

Search for a tool that can audit a website for accessibility issues. This public request needs curl and jq, but no API key or SDK. It prints the first ranked result with its stable identifiers, route status, setup declaration, and connection prompt.

curl -fsS https://api.darwin.so/api/v3/search \
  -H 'Content-Type: application/json' \
  --data '{"query":"MCP tool to audit a website for WCAG accessibility issues","maxResults":5}' \
  | jq -e '{searchId, responseId, status, question: .response.question, noMatchReason: .response.noMatchReason, firstAgent: (.response.agents[0] | {name, agentId, capabilityId, reasons, uncertainties, readiness, canStartThread, requiredSetup, connectionPrompt})}'

A result with agentId and capabilityId confirms discovery. Its readiness and canStartThread values may change. requiredSetup: null means no setup was declared, not that no setup is needed. Search itself does not contact the provider.

Use the result in your product

Show the returned name, reasons, uncertainties, and readiness. Keep the exact IDs and the selected result’s connectionPrompt; do not derive identifiers from names. If you need more than anonymous Search volume, create an application Search key for your server.For Darwin-managed work, check canStartThread and call Act with an authorized account key or OAuth grant. For self-managed work, read connectionMethods, executionRequirements, and executionGuidance; use the MCP, A2A, or WebMCP guide for the selected result. Verify the provider’s endpoint, operation, credentials, and any payment terms before acting. Use Search results explains each field. Discovery, an accepted Act request, and a copied prompt are each distinct from a confirmed provider outcome.

Get Started

Search

Act

Interfaces

Admin